Policy Orchestrator
Centralized policy check and routing for the AIMM agentic trading system: decide whether a given ticker is allowed for a user, with reason and policy ID.
SKILL.md declares no external API dependencies, acts as a hard policy gate (blocks violating trades, defaults to paper routing), and points to local config/policy_loader.py — narrow external impact. Deducted for: unspecified permission scope, no user confirmation mechanism, no rollback path, and hot-reloadable rule content not visible in evidence.
Static read only: interface contract, JSON output shape, and query example are internally consistent, but the referenced policy_loader.py and policy_types.py are absent from evidence, so dependency availability is unconfirmed. Deducted for: no edge-case handling or failure-feedback documentation.
Clear scenario (user policy check / asset approval routing), precise trigger semantics (ticker/action=check), purely local rules reachable from mainland China. Deducted for: no declared non-fit boundaries, no off-domain behavior, no Chinese-language support statement.
AGPL-3.0 license is clear, persona.md and SKILL.md are readable layers, and the repo has SECURITY.md plus a test system. Deducted for: no skill-level version, changelog, parameter-stability commitment, known-limitations disclosure, or explicit maintenance ownership.
Static calibration caps this at 7: the claimed core task (allowed/blocked + route + reason + policy_id) has a directly usable output format, but no representative output evidence and insufficient argument for marginal value over manually querying policy rules — deducted.
The repo has a CI workflow and pytest suite, but no test covering policy_orchestrator key paths appears in the evidence files. Deducted for: behavioral claims lacking independently reproducible material beyond limited auditable source.
- This is a static source review; no code was executed. The referenced policy_loader.py / policy_types.py could not be verified from the provided evidence.
- This skill is an execution-gating component for trading; audit the policy rule implementation and hot-reload safety yourself before production use.
- Publisher identity is not verified by the FollowSkills registry and is treated as unknown.
- No skill-level version or changelog; behavior may change across updates.
- No Chinese-language support statement; while rule dependencies are local, the broader platform depends on overseas LLM APIs (OpenAI/Atlas Cloud).
What does this skill do, and when should you use it?
Policy Orchestrator is one of 14 skills bundled in olaxbt/ai-market-maker (AIMM, an open-source agentic AI hedge fund OS), located at src/agents/governance/policy_orchestrator/. It reads runtime policy definitions from local config files, runs a policy check for a given ticker and action, and returns an allowed/blocked decision with route, reason, and policy ID. It has no external API dependencies and supports hot-reloadable rules. Adopt it if you need to enforce centralized trading policy (which tickers are permitted, per user) inside an agentic trading pipeline.
Exposes evaluate(signal, user_config) returning a route decision plus policy check; supports direct queries such as "Is BTC allowed for this user?" and "What are the current policy rules?"; serves a query interface /policy_orchestrator?ticker=BTC/USDT&action=check returning allowed/blocked, route, reason, and policy_id; loads hot-reloadable rules from config/policy_loader.py and the policy schema from config/policy_types.py; makes no external API calls.
- A crypto hedge fund team that must verify a ticker (e.g. BTC/USDT) against current policy rules before execution.
- An operator of a multi-user trading platform needing per-user restrictions on tradable assets.
- A governance/compliance reviewer who wants to list the currently active policy rules on demand.
- A developer building a pre-execution policy gate inside the AIMM multi-agent trading stack.
What are this skill's strengths and limitations?
- No external API dependencies; policy checks are fully local.
- Rules hot-reload via policy_loader, so policy changes need no restart.
- Results include reason and policy_id, making decisions traceable.
- Simple surface: one function plus one query endpoint.
- Documentation is minimal — just the SKILL.md capability list, with no test evidence or example outputs.
- Policy rule content must be maintained yourself in config; the skill provides no default rules guidance.
- As one of 14 skills in a monorepo, its coupling to full-platform installation is unclear.
- How the /policy_orchestrator endpoint is deployed (e.g. exposed via the API service) is not documented.
How do you install this skill?
The skill ships with the repository: git clone https://github.com/olaxbt/ai-market-maker.git, then follow README setup (pip install uv; install TA-Lib; uv sync --extra dev). The skill file is at src/agents/governance/policy_orchestrator/SKILL.md. The repo also supports OpenClaw installation (claw install https://github.com/olaxbt/ai-market-maker or claw skill install ./openclaw), but those commands install the whole collection; standalone installation of just this skill is not documented.
How do you use this skill?
In a configured environment, call evaluate(signal, user_config), or query /policy_orchestrator?ticker=BTC/USDT&action=check to get allowed/blocked + route + reason + policy_id. You can also ask an agent directly, e.g. "Is BTC allowed for this user?". Note: the README covers installing and running the whole AIMM platform (Docker stack, testnet trading); the minimal environment for running this skill alone is not documented.